Hidden Non-Detects Inflate Stream Pesticide Rates Up to 90-Fold
Key result: In the US national water-quality record, 39,971 insecticide results from stream samples carry a numeric value but are coded elsewhere as not detected. Read at face value, they inflate detection rates up to 90-fold: permethrin appears in 39.3 percent of samples, but in 1.3 percent once they are handled correctly.
Why this audit was redone
The first version of this report, published on 2 October 2026, tested whether national detection rates for pest control insecticides survive the removal of any single monitoring programme. Several did not, and we attributed the swings to programme dominance. Building a follow-up analysis against federal aquatic life benchmarks, we found the real cause: thousands of results in which a laboratory recorded a number, usually its detection limit, while marking the sample as not detected somewhere other than the field most analyses read.
Every rate in that first version, and in our earlier report on insecticides in US streams, read those results as detections. This version replaces it. The correction is set out in full at the end.
Three ways a non-detect hides in the record
The Water Quality Portal, run by the US Geological Survey, the Environmental Protection Agency and the National Water Quality Monitoring Council, standardises how results are stored. It has a field for the detection condition, where “Not Detected” belongs. Most analyses, including ours, read that field. Three groups of contributors record non-detects another way.
| Mechanism | Results affected | Main contributors | How it is coded |
|---|---|---|---|
| Florida qualifier flag | 27,529 | Florida Department of Environmental Protection districts and 13 Florida programmes in all | Detection limit entered as the value; “**U**” written at the start of the comment text |
| Programmes that never report a non-detect | 9,661 | Indiana (8,575), South Dakota (623), a volunteer watershed group (360), Missouri (103) | Every result coded “Present Below Quantification Limit”, none “Not Detected” |
| WQX “U” qualifier with a value | 2,781 | Washington State Department of Agriculture (2,650), EPA Region 7 (130) | Value entered; qualifier code “U”, defined by EPA’s WQX standard as Not Detected |
| Total | 39,971 |
Source: Water Quality Portal, stream results for 14 insecticides from 1 January 2015, retrieved 25 September to 3 October 2026. Author classification. Florida’s flag is defined in Florida Administrative Code rule 62-160.700, Table 1; the WQX code in EPA’s ResultMeasureQualifier domain list.
Florida’s flag is the largest. Florida’s quality-assurance rule defines the qualifier exactly: “U Indicates that the compound was analyzed for but not detected… The value associated with the qualifier shall be the laboratory method detection limit.” Florida laboratories follow the rule. When the results reach the national portal, the qualifier lands in the comment text and the detection-condition field is left blank, so the detection limit looks like a measured concentration.
The never-non-detect programmes are the least certain. Indiana reported 3,351 of its 3,356 chlorpyrifos results from 2015 to 2026, and every one of its bifenthrin and permethrin results, as “Present Below Quantification Limit”, with not one non-detect. A compound found in every sample for eleven years while nearly every other programme finds it rarely is not plausible, so we treat these results as uninformative and exclude them. Counting them as non-detects instead changes no corrected rate by more than 1.7 percentage points.
The WQX qualifier is the most straightforward. EPA’s standard defines “U” as “Not Detected: The analyte was analyzed for, but was not detected”. A value accompanying it is a reporting limit.
What the corrected rates look like
| Compound | Results | Naive rate | Corrected rate | Lower bound | Inflation | Corrected site rate |
|---|---|---|---|---|---|---|
| Imidacloprid | 36,081 | 53.2% | 41.6% | 40.0% | 1.3x | 59.8% |
| Clothianidin | 18,066 | 52.4% | 22.1% | 20.4% | 2.4x | 32.0% |
| Thiamethoxam | 17,232 | 51.8% | 17.6% | 16.2% | 2.9x | 24.8% |
| Deltamethrin | 3,100 | 49.7% | 0.6% | 0.5% | 85.1x | 2.8% |
| Chlorfenapyr | 2,070 | 48.7% | 1.2% | 1.2% | 40.3x | 2.2% |
| Dinotefuran | 14,917 | 44.0% | 12.5% | 12.3% | 3.5x | 25.5% |
| Permethrin | 5,457 | 39.3% | 1.3% | 1.2% | 30.4x | 6.1% |
| Fipronil | 25,042 | 27.2% | 17.0% | 16.1% | 1.6x | 31.9% |
| Chlorpyrifos | 37,298 | 16.5% | 2.7% | 2.5% | 6.1x | 12.5% |
| Cyfluthrin | 12,637 | 13.3% | 0.7% | 0.7% | 18.3x | 5.2% |
| Esfenvalerate | 11,651 | 11.5% | 0.3% | 0.3% | 37.3x | 2.6% |
| Pyriproxyfen | 12,486 | 10.8% | 0.1% | 0.1% | 89.7x | 0.9% |
| Bifenthrin | 21,686 | 10.7% | 2.6% | 2.6% | 4.1x | 13.1% |
| Indoxacarb | 11,367 | 0.1% | 0.1% | 0.1% | 1.0x | 1.4% |
Source: Water Quality Portal, US stream sites, water results from 1 January 2015. Naive rate counts any numeric result as a detection. Corrected rate applies all three corrections and excludes programmes that never report a non-detect. Lower bound counts those programmes’ results as non-detects instead. Site rate is the share of monitored sites with at least one corrected detection. Author calculation.
What it means. The pyrethroids are hit hardest. Deltamethrin falls from 49.7 to 0.6 percent, permethrin from 39.3 to 1.3, esfenvalerate from 11.5 to 0.3. These compounds are rarely found in the dissolved water samples most programmes collect, and much of their apparent presence was detection limits. The neonicotinoids are still common after correction: imidacloprid at 41.6 percent of samples, clothianidin 22.1, thiamethoxam 17.6, dinotefuran 12.5. Fipronil is found in 17.0 percent. The ranking of what is actually common in US streams is now clear: neonicotinoids and fipronil, not pyrethroids.
What limits it. We can only correct the conventions we found. A programme that enters detection limits as values with no flag of any kind would still be read as detections; the agreement between the corrected and lower-bound rates suggests no large convention remains, but it cannot prove it.
Does any single programme still drive a rate?
| Compound | Programmes | Largest programme’s share | Range with any one programme removed |
|---|---|---|---|
| Imidacloprid | 71 | 11.4% | 39.9% to 44.2% |
| Clothianidin | 42 | 23.6% | 19.4% to 23.8% |
| Thiamethoxam | 41 | 24.9% | 14.3% to 20.6% |
| Deltamethrin | 16 | 44.2% | 0.3% to 1.0% |
| Chlorfenapyr | 18 | 36.3% | 0.9% to 1.9% |
| Dinotefuran | 35 | 26.7% | 10.9% to 17.0% |
| Permethrin | 24 | 24.4% | 1.2% to 1.5% |
| Fipronil | 70 | 10.3% | 15.7% to 18.9% |
| Chlorpyrifos | 82 | 21.9% | 2.3% to 3.1% |
| Cyfluthrin | 39 | 60.0% | 0.1% to 1.8% |
| Esfenvalerate | 19 | 64.1% | 0.1% to 0.9% |
| Pyriproxyfen | 59 | 11.9% | 0.1% to 0.1% |
| Bifenthrin | 63 | 34.7% | 1.0% to 4.0% |
| Indoxacarb | 48 | 17.8% | 0.1% to 0.2% |
Source: as above, corrected rule. Author calculation.
What it means. Once non-detects are read correctly, no single programme moves any compound’s rate by more than about 5 percentage points, and for most compounds by under 2. The large programme effects reported in the first version of this audit, such as an 85 percent deltamethrin rate when one programme was removed, were produced by the hidden non-detects, not by any programme’s sampling choices.
What limits it. Programmes still choose where to sample, so these rates describe the monitored streams rather than all US streams.
Our own errors, corrected
| Figure | As first published | Corrected |
|---|---|---|
| Imidacloprid, share of stream samples | 53.3% | 41.6% |
| Fipronil | 27.4% | 17.0% |
| Chlorpyrifos | 17.0% | 2.7% |
| Bifenthrin | 11.0% | 2.6% |
| Permethrin | 39.4% | 1.3% |
| Deltamethrin | 49.9% | 0.6% |
| Thiamethoxam | 51.8% | 17.6% |
| Dinotefuran | 44.0% | 12.5% |
First-published figures from our streams report and the first version of this audit.
Our streams report, our report on fipronil breakdown products and this audit have been corrected. The streams report’s conclusion that a 2024 to 2025 surge in fipronil and bifenthrin detections was a laboratory artifact is replaced: on the corrected rule there is no surge. The apparent jump came from Indiana’s results, which begin in 2024, and from Florida’s flagged values.
Three tests anyone can run
1. Read the comment field
Method. Filter results with a blank detection condition and a numeric value. Count how many have comment text beginning “**U”.
Result. 27,529 such results across 14 insecticides since 2015, all from Florida programmes.
2. Read the qualifier field
Method. Count numeric results whose MeasureQualifierCode contains the code U, including combinations such as “J;U”.
Result. 2,781 results, 2,650 of them from the Washington State Department of Agriculture.
3. Check every programme for non-detects
Method. For each compound, flag any programme with at least 50 results, no non-detects, and at least half its results coded “Present Below Quantification Limit”.
Result. Indiana for ten compounds; South Dakota, a volunteer watershed association and Missouri for one to three each.
Related: the corrected detections compared with EPA’s benchmarks are in insecticides in US streams against EPA’s benchmarks.
Download the data
The tables in this report are published as CSV files under a Creative Commons Attribution 4.0 licence. You may republish, remix and build on them, including commercially, with attribution to The Pest Control Brief and a link to this page.
- Naive and corrected detection rates, inflation factors and programme robustness for 14 insecticides (CSV, 14 rows)
- Results recorded as values but coded as not detected, by mechanism and programme (CSV, 20 rows)
How to cite this report
The Pest Control Brief, “The National Pesticide Monitoring Record, Audited: Hidden Non-Detects Inflate Insecticide Detection Rates Up to 90-Fold”, 2 October 2026. https://thepestcontrolbrief.com/national-pesticide-monitoring-reliability/
References
- Water Quality Portal. Result search, narrowResult profile, siteType=Stream, from 1 January 2015, for imidacloprid, clothianidin, thiamethoxam, dinotefuran, fipronil, bifenthrin, permethrin, deltamethrin, cyfluthrin, esfenvalerate, chlorpyrifos, chlorfenapyr, indoxacarb and pyriproxyfen. waterqualitydata.us. Retrieved 25 September to 3 October 2026.
- Florida Department of Environmental Protection. Chapter 62-160, Florida Administrative Code, rule 62-160.700, Table 1: Data Qualifier Codes. floridadep.gov 62-160 Table 1 (PDF). Accessed 3 October 2026.
- US Environmental Protection Agency. WQX domain values: ResultMeasureQualifier. cdx.epa.gov ResultMeasureQualifier.CSV. Accessed 3 October 2026.
How the audit was done
Data. Every stream-site water result since 1 January 2015 for 14 insecticides used in pest control, 229,090 results after excluding sediment, tissue, quality-control recoveries, passive samplers, and results without a usable unit or date.
Corrected rule. A result is a non-detect if its detection condition says so, if its qualifier code contains U, or if its comment text begins with a Florida U flag. A result is a detection if its condition is “Present Below Quantification Limit” or “Trace”, or if it carries a positive numeric value without those markers. Programmes with at least 50 results for a compound, no non-detects and at least half below-quantification results are excluded for that compound.
Robustness. Each compound’s corrected rate recomputed with every programme removed in turn.
Limitations. Conventions not marked anywhere in the record cannot be detected. Detection depends on each laboratory’s reporting limit, which varies. Monitored streams are not a random sample of US streams.
Corrections. This report was first published on 2 October 2026 with rates computed under a rule that read hidden non-detects as detections, and with programme-sensitivity findings that those non-detects produced. It was rewritten on 3 October 2026. All rates, the robustness table and the conclusions changed; the first version’s figures are listed in the section above.
Last reviewed: 3 October 2026.
For researchers and reporters using water-quality data
Most quotable statistics
- 39,971 insecticide results from US stream samples since 2015 carry a numeric value but are coded elsewhere as not detected.
- Read naively, permethrin is detected in 39.3 percent of US stream samples; corrected, 1.3 percent.
- Deltamethrin’s naive detection rate of 49.7 percent falls to 0.6 percent when non-detects are read correctly, an 85-fold inflation.
- 27,529 of the hidden non-detects come from Florida programmes, which follow a state rule that enters the detection limit as the value.
- After correction, imidacloprid remains the most commonly detected insecticide, in 41.6 percent of US stream samples.
Advice for anyone analysing the portal
Never treat a numeric value as a detection without checking the qualifier code and the comment text, and check each contributing programme for whether it ever reports a non-detect.
Press summary
Tens of thousands of non-detects are hidden in plain sight in the US national water-quality database, and reading them at face value inflates pesticide detection rates as much as 90-fold, according to an audit by The Pest Control Brief. Since 2015, 39,971 insecticide results from stream samples carry a number that is actually a laboratory’s detection limit, flagged as not detected in a comment, a qualifier code or by programme convention. Corrected, permethrin is found in 1.3 percent of samples rather than 39.3 percent and deltamethrin in 0.6 rather than 49.7. Neonicotinoids remain common: imidacloprid is found in 41.6 percent. The Pest Control Brief has corrected three of its own earlier reports that made the same error.
Headlines editors could use
- The hidden non-detects that make pesticides look 90 times more common
- How a Florida data rule inflated national pesticide statistics
- Permethrin in 39 percent of US streams? Try 1.3 percent
- Neonicotinoids, not pyrethroids, are what US streams actually contain
Questions about national monitoring data
How common are pesticides in US streams?
Of 14 insecticides used in pest control, corrected detection rates in US stream water samples since 2015 range from 41.6 percent for imidacloprid to 0.1 percent for indoxacarb and pyriproxyfen.
Why do some analyses report much higher detection rates?
Because some programmes store a detection limit as a numeric value and mark the sample as not detected only in a comment, a qualifier code, or not at all. Reading the value as a detection inflates the rate.
What is a “U” qualifier?
A code meaning the analyte was analysed for but not detected. In Florida’s rule the value reported alongside it is the laboratory’s method detection limit.
Which pesticides are most common in US streams after correction?
Neonicotinoids: imidacloprid 41.6 percent, clothianidin 22.1, thiamethoxam 17.6 and dinotefuran 12.5 percent of samples, followed by fipronil at 17.0 percent.
Are pyrethroids rarely found in streams?
Rarely in stream water samples: bifenthrin 2.6 percent, permethrin 1.3, deltamethrin 0.6. Pyrethroids bind to sediment, which most water sampling does not measure.
Is the Water Quality Portal unreliable?
No. The information is in the record; it is in fields many analyses do not read. Read correctly, the data are consistent across programmes.
Did The Pest Control Brief make this error?
Yes. Our streams report, our fipronil breakdown products report and the first version of this audit all read the hidden non-detects as detections. All three have been corrected.
Can I check this?
Yes. The three tests above use only fields present in every Water Quality Portal download.
